The boundaries section says restore is file-content atomic per file, not a full workspace transaction. That is an honest way to put it, and it is the line I got stuck on.
If a multi-file revert fails partway, the tree ends up in a state that never existed. The safety checkpoint means the work is still recoverable, so that part is fine. What I cannot tell is whether the record notices.
Does the operation log write the intended path set before applying, so a later reader can tell a completed restore from a half-applied one? Or does the entry describe only what actually landed?
The boundaries section says restore is file-content atomic per file, not a full workspace transaction. That is an honest way to put it, and it is the line I got stuck on.
If a multi-file revert fails partway, the tree ends up in a state that never existed. The safety checkpoint means the work is still recoverable, so that part is fine. What I cannot tell is whether the record notices.
Does the operation log write the intended path set before applying, so a later reader can tell a completed restore from a half-applied one? Or does the entry describe only what actually landed?